Integrating UAV-Derived Diameter Estimations and Machine Learning for Precision Cabbage Yield Mapping.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Conventional cabbage yield estimation is labor-intensive and time-consuming.
• Unmanned Aerial Vehicle (UAV) imagery offers a more efficient and spatially aware approach.
• Assessing spatial variability aids resource allocation and variable rate applications.

Purpose of the Study:

• To develop a non-destructive method for estimating cabbage head diameter using UAV imagery.
• To predict individual cabbage head fresh weight using estimated diameters and environmental data.
• To evaluate the performance of deep learning and machine learning models for precision agriculture.

Main Methods:

• Individual cabbage head diameters were estimated using YOLOv8s-pose and YOLOv11s-pose deep learning models with high-resolution RGB UAV imagery.
• Climatic variables (temperature, rainfall) and canopy reflectance indices (NDVI, NDRE, CIg) were extracted from multispectral UAV imagery.
• Machine learning models, including CatBoost, were employed to predict fresh weight using diameter estimations and environmental data.

Main Results:

• YOLOv11s-pose demonstrated high accuracy in diameter estimation with a mean relative error of 4.6% and mAP of 98.5%.
• CatBoost achieved the lowest Mean Squared Error (0.025 kg/cabbage) and highest R² (0.89) for fresh weight prediction.
• The integrated AI framework significantly improved non-invasive yield estimation accuracy.

Conclusions:

• Deep learning-based pose estimation with UAV imagery provides accurate non-destructive cabbage diameter measurement.
• An AI-powered framework combining diameter, climatic, and spectral data enhances precision yield prediction in cabbage farming.
• This approach supports efficient resource management and supply chain optimization in agriculture.